Discussion Forum

Hello!

I know that neural network techniques can be used for both classification and regression. I was wondering if the use of an activation function at the output layer of a NN depends on whether we are doing classification or regression?

I don't think so, but I got that impression since it seemed to me that when we were using MLPs, we used activation functions on every single layer including the output layer. Yet when using CNNs etc, I realized that sometimes the output layer doesn't have an activation function. I don't really understand why and would appreciate an explanation as to when/why to use an activation function on the last layer, and if it really matters.

Thank you for your time,
Johnny